Balancing Technology and Copyright: Clarifying IP Boundaries for Animation Industry in the AIGC Era

As generativeAI technologies gain widespread adoption in animation and shortvideo sectors, the AI mandrama segment has experienced explosive growth. Industry analysts project China’s AI mandrama market size will reach RMB 24.36 billion in 2026. While AI drastically cuts production costs and lead times, copyright conflicts over trainingdata sourcing, ownership of AIgenerated outputs and commercial exploitation have surfaced intensively. The copyrightinfringement case of domestic Chinese animation Wushan Wuxing (Fog Hill of Five Elements) has become an industry landmark, sparking wideranging discussions among industrial and legal circles over the core principle of “authorisation first, utilisation later”.

On August 21, 2026, Liu Dao Wu Yu Studio, copyrightholder of the celebrated domestic animation Fog Hill of Five Elements, issued a formal public statement. It accused a Bilibili content creator of scraping original animation frames and art backgrounds without permission to train a selfbuilt AI model, which was then used to produce and publicly distribute AIanimated work Da Ai Xian Zun. Comparative checks revealed numerous scenes and art styles in the infringing AI work bore striking resemblance to the original animation, with segments amounting to direct copying, allegedly violating rights of reproduction, adaptation and informationnetwork communication. The copyright holder demanded full removal of all relevant source materials and training parameters from the AI model, complete takedown of infringing AI works across platforms and a formal public apology. The involved content creator later published an apology and removed all related videos, yet systemic industry risks exposed by the incident persist.

From a technical perspective, multimodal largescale models carry the inherent risk of “deep memorization and content regurgitation”. Without directly duplicating original files, pretrained AI models may reconstruct artstyles, compositions and narrative features of copyrighted works via parameterweight memorization. This raises core legal questions on the input side: whether scraping copyrighted creative works for commercial AImodel training qualifies as legitimate practice. Legal and intellectualproperty experts have reached a clear consensus: technological neutrality cannot serve as justification for misappropriating original creations. Commercial AI training activities must adhere to the rule of “authorisation first, utilisation later”. Animation, comic and literary IPs cannot be arbitrarily scraped as training corpora. The Interim Measures for the Administration of GenerativeAI Services stipulate that commerciallyoriented AI training must draw on legallysourced materials, and unauthorised use of copyrighted content for training triggers corresponding infringement liabilities.

Copyright challenges also exist on the output side. The AImandrama sector faces a widespread imbalance: infringement outpaces copyright enforcement. AI tools lower barriers for mass content production. Some producers carry out “rewashstyle” adaptation of online novels and comic IPs to launch content and capture traffic revenue rapidly. Meanwhile, original rightholders face complex, highcost evidencegathering procedures for litigation. In many instances, infringing parties have already realised commercial gains by the time enforcement actions take effect. Regarding ownership confirmation, Chinese judicial practice has established consistent criteria. Pure oneclick AI outputs lacking meaningful human creative intervention do not qualify as copyrighteligible works. For commercial use and copyright protection, human creative intervention should account for no less than 35 % of finished works, covering storyboard design, characterdetail development, iterative parameter tuning and content screening. Creators must preserve complete evidence including revision drafts and promptiteration logs. Insufficient human contribution prevents AI mandramas from realising IP capitalisation, blocking derivativeproduct development and copyright licensing business models.

Regulatory authorities in China are building comprehensive institutional frameworks to bring unregulated AImandrama growth under oversight. The National Radio and Television Administration has incorporated AIGCanimated microdramas into a tiered filingandreview regime. Review authorities are assigned according to project investment scale and thematic risk levels. Highbudget and sensitivetheme productions are reviewed centrally, while tailored filing pathways are available for smallandmedium creators. Unfiled legacy noncompliant content will be removed from online platforms. Industry representatives including nationalcommittee member Zhou Yuan have proposed settingup fasttrack disputeresolution channels for AIgeneratedcontent infringements, implementing the “verifyandtakedown” complaint mechanism, introducing penalty scales linked to infringingparty revenue, strengthening sanctions against massscale infringements and reinforcing platform accountability to safeguard creators’ legitimate rights.

Industry stakeholders widely acknowledge that AI serves as an efficiencyenhancing creative tool rather than a loophole to evade copyright obligations. While embracing AIdriven productivity gains, the animation sector must uphold intellectualproperty standards. A healthy ecosystem requires legallylicensed training datasets, evidencebacked ownership confirmation for AI outputs, and coordinated governance between regulators and platforms, to achieve dynamic balance between technological innovation and originalcontent protection.